Web Search
AI models have knowledge cutoffs because they can't learn new information after training. To access current information or web results, you can use the Web Search tool.
Web Search
AI models have knowledge cutoffs because they can't learn new information after training. To access current information or web results, you can use the Web Search tool.
The more context and details you add, the better your response because the model understands precisely what you expect. Do not miss our Prompt Engineering Guide to learn how to write great prompts.
Web search solves a core technical limitation of AI models. Large Language Models go through two phases: training (when they're "built") and then deployment (when you use them). Once training is complete, the model's knowledge is frozen at that cutoff date and can't be updated. This means even the newest models become outdated the moment they're released.
The web search tool bridges this gap in two steps:
- Search: A query is generated and searches the internet for relevant results
- Context: Those results get sent to the AI along with your prompt to generate an informed answer
Perfect for:
- Gathering current information on any topic
- Getting real-time data and recent developments
- Searching specific websites (just include the URL in your prompt)
Want to understand more about how AI training works? Check out our guide about how AI works.
Selecting web search
To use Web search, open a new chat and select a model that supports web search. You’ll see web search availability indicated by the icon in the model selector if the web search icon is greyed out, that model doesn’t support web search.
To activate web search, click the + button in the input bar, then select Web search from the menu.
Once activated, the Web search button appears next to the + button and highlights to confirm it's active. You can now search the internet or access websites.
Using web search
Web search automatically triggers when the AI detects that your prompt requires information beyond its training cutoff date.
Your then prompt gets reformatted into an optimized search query, then our search model finds relevant results across the web. Once the search is complete, the model analyzes findings from multiple websites and synthesizes them into a comprehensive answer for you.
The whole process happens seamlessly in the background, so you get current information without any extra steps on your end.
Inspecting sources
When you want to see what websites the AI used to write the response, click on Searched for "your search query" to view the complete list of websites that were analyzed.
On this view, you can see the citations and search results from your web search.
When you hover over a citation, it highlights exactly which paragraph quoted that specific website. Below the main response, you'll find all the other search results that were analyzed during the search but didn't make it into the final answer.
This gives you full transparency into both what sources were used and what additional information was considered but not included.
Opening specific URLs
While web search finds relevant pages across the internet, sometimes you need to access a specific URL directly. The open_url tool retrieves and renders the content of a webpage you specify.
When to use open_url vs web search
| Use case | Tool |
|---|---|
| Research a topic with unknown sources | Web search |
| Access a specific page you already know | open_url |
| Get current news or recent developments | Web search |
| Read an article someone shared with you | open_url |
| Compare information across multiple sites | Web search |
How it works
When you share a URL with the AI, the open_url tool fetches the page content and converts it to text that the model can analyze. This is useful for:
- Summarizing articles or documentation
- Extracting specific information from a known page
- Analyzing content from links shared in conversations
Usage limits
The open_url tool has a limit of 3 calls per turn. If you need to access more than 3 URLs, split your requests across multiple messages.
Best practices
- Be specific about what you need: Instead of just pasting a URL, tell the AI what you're looking for. For example: "Summarize the key points from this article: [URL]"
- Use web search for discovery: If you're not sure which page has the information you need, start with web search to find relevant sources
- Batch related URLs: If you need content from multiple pages, include up to 3 URLs in a single message to stay within the limit